A Coding Font Odyssey: Hangul Comes to JetBrains Mono
Finding a coding font you actually like is harder than it sounds. Even D2Coding left something to be desired. Then I found a project that bolted Hangul and Nerd Font onto JetBrains Mono.
Thoughts on software engineering, AI, and technology.
Finding a coding font you actually like is harder than it sounds. Even D2Coding left something to be desired. Then I found a project that bolted Hangul and Nerd Font onto JetBrains Mono.
It started with a dumb Sunday-afternoon impulse to write my own name on my GitHub profile. While I was pushing dithering around on a tiny 53×7 grid with five shades of grey, I stumbled into two surprisingly weird corners of GitHub's contribution-graph algorithm.
Customers want a cheap, fast BMS that just barely clears the regulatory minimum, while academia keeps pushing safety from the other side. Follow the currents that markets, regulation, and the economics of battery fires create between them, and you can see why high-safety BMS is quietly becoming a survival condition.
While dissecting STM32CubeMX2, one thing made me pause. Every GUI action maps 1:1 to a CLI command, and even the GUI is just one client of an HTTP backend. Here's why that structure matters in the age of AI agents.
Embedded, automotive, firmware — the slow-moving corners of tech where new patterns always arrive a beat late. Modern software's mindset is finally showing up here too. I dissected STM32CubeMX2 disk-by-disk and pulled out the signals.
Parsing Ultima Online's binary packets, synchronizing state, and deciding when to use an LLM. Putting an AI agent into a living MMORPG requires solving more than you'd think.
Ultima Online took my youth in 1998. Staring at an empty private server, the thought of 'what if AI lived in this world' turned into a project.
Stanford researchers built a virtual village called 'Smallville' where 25 AI agents spontaneously formed a society. Reading that paper, what came to mind was the time I spent in Britannia back in 1998.
At a small embedded systems company, I'm watching a paradox unfold: the more AI boosts senior productivity, the less room there is for juniors to stand.
After four years of teaching software at a university, I've found it increasingly hard to explain to students why they should learn to code. I found part of an answer in an essay by the creator of htmx.
I was laid up with a stomach bug over the holiday, and when I finally dragged myself to my desk, Claude Code's servers were dead. And I found myself not even thinking to try Codex — even though it was right there.
I wanted to watch multi-agents work with my own eyes. Starting from the single question 'What are you doing right now?', it took two days to build a real-time dashboard and release it as open source.
If SSH works fine but the screen is frozen, a memory region conflict between efifb and vmwgfx may be the cause. The root cause and fix for this ARM64-only issue.
From a Vercel template to a fully custom shadcn/ui blog — the story of building a blog without typing a single line of code myself.
I've tried Teams, Slack, and Mattermost, and always found something lacking. Workplace chat where AI agents participate as colleagues could change that entirely.
The first thing that annoyed me when I got a MacBook was Korean/English input switching. I didn't know a single line of Swift, but I built an app with vibe coding and got it onto Homebrew.
Starting from a story about reviving a 25-year-old kernel driver, this post explores the sense of efficacy that LLMs give us — the ability to finally build things that were always necessary but that nobody ever touched.